Customer-First Direct Callback

What it does

Customer-first direct callback calls the customer first when their callback request reaches the front of the queue, then connects them to a representative once they've answered and passed any verification. The classic model does the opposite — it ties up a representative while the system dials out and waits through voicemail or an unanswered ring. Calling the customer first means representatives only get engaged on calls that actually connect, so idle time drops. Behaviour is driven by callback profiles you attach to a queue's overflow action.

Key facts

Enablement is irreversible — once you turn on the Copilot AI agent for callback management you can't turn it off, and the first profile you create is automatically applied to any queue already using direct callback as an overflow action.
It provisions a Callback Agent in Copilot Studio; you can't enable the feature until that agent provisions cleanly and you've created at least one callback profile.
Every callback runs through an AI agent that can verify the customer, collect information, present context, and route or escalate to a representative — this consumes Copilot / Microsoft service credits under usage-based billing.
Prerequisites: a provisioned Copilot Studio agent, phone numbers with outbound calling enabled, and any verification/collection agents you plan to use.
Profiles cover the whole flow: alternate-number capture, initial offer message, repeat interval, callback offering window (with time zone taken from queue operating hours, UTC if none set), duplicate prevention (per-queue or across all queues), and queue allocation limits that pause offers when full and resume as capacity frees up.
Dialling supports the originally dialled number (with a fallback for numbers that can't do outbound) or a dedicated outbound number, plus fallback actions (prompt-and-hang-up or prompt-and-escalate) when the agent can't engage.
The rules engine dynamically delays new callbacks when the recent average wait exceeds a target you set (e.g. five minutes over the last 30), shrinking the delay as staffing recovers — this stops customers answering a callback only to sit on hold for a representative.
Retry behaviour is configurable per trigger (no answer, failed, answering machine) with a minimum retry wait of five minutes.
One callback profile per overflow condition per queue.

When to use / skip

Use it where representative idle time on outbound callbacks is a real cost — high callback volumes, spiky demand, or queues where classic direct callback ties agents up on calls that never connect. The dynamic wait-time rules make it a good fit for contact centres that want callback pacing to track staffing automatically.

Skip it if your callback volumes are low, or if you're not ready for the AI-agent dependency and its credit consumption. And think hard before enabling in production for the first time — you can't undo it, and it immediately reshapes any queue already using direct callback overflow.

Configuration decisions

Which queues move to customer-first callback, given enablement is one-way and retrofits existing direct-callback queues on day one.
Duplicate handling: check within the current queue only or across all queues, and whether a duplicate ends the call or keeps the customer in the queue.
Callback number strategy: reuse the dialled-in number (and its fallback) or always use a dedicated outbound number.
Target average wait time for the pacing rules, and whether callbacks are allowed outside queue operating hours.
Retry counts and wait durations per trigger, and the fallback action when the AI agent can't engage.
Which Copilot Studio agent handles verification and information collection, and what topics it needs.

Gotchas

There's no off switch. Enablement is permanent — plan the rollout as a one-way door and test in a non-production environment first.
Editing or deleting the alternate-number AI agent breaks callback. Microsoft explicitly warns against touching the agent configured for taking alternate numbers.
Credits are consumed per callback. Every callback runs an AI agent, so the same licensing and usage rules as other AI agents apply — budget for it before you scale.
First profile applies automatically to existing direct-callback queues. If you've got queues already using direct callback as overflow, they inherit your first profile the moment you finish setup, whether you intended that or not.
Queue allocation is a hard cap. When a queue hits its active-callback limit the system stops offering callbacks and pushes customers back into the standard experience — size the limit against real volumes.

Consultant notes

The irreversibility is the thing clients underestimate in scoping. Get sign-off that customer-first is the target model before you flip it, because there's no rollback and the first profile immediately rewrites the behaviour of any queue already doing direct-callback overflow. Demo it in a sandbox and walk stakeholders through that day-one effect.
The AI-agent dependency drags Copilot Studio provisioning and credit consumption into what clients think of as a telephony feature. Raise credits and the Copilot Studio agent as scoping dependencies early — procurement and the platform team both need to be in the room, not just the contact-centre lead.
The dynamic wait-time pacing is genuinely useful but needs a realistic target agreed with operations. Set it too tight and you throttle callbacks unnecessarily; too loose and customers still answer to dead air. Tune it against actual queue data during UAT rather than guessing.
Watch the interaction with existing callback and deduplication config. This feature carries its own duplicate-prevention settings, so make sure you're not layering conflicting logic against the classic callback and dedup behaviour already in place.

Worth revisiting after the next release wave — profile options and the AI-agent billing model are new and likely to move.